Информатика ва ахборот технологиялари

3-секция.Информатика ва ахборот технологиялари.

DEEP LEARNING IN CELL IMAGE ANALYSIS
¹Muhammad al-Xorazmiy nomli Toshkent Axborot texnologiyalari Universiteti Samarqand filiali Dasturiy injiniring kafedrasi katta o`qituvchisi
Safarova G.T.
<sup>2</sup>Muhammad al-Xorazmiy nomli Toshkent Axborot texnologiyalari Universiteti Samarqand filiali Dasturiy injiniring kafedrasi katta o`qituvchisi
Maxkamova D.A.
³Muhammad al-Xorazmiy nomli Toshkent Axborot texnologiyalari Universiteti Samarqand filiali Dasturiy injiniring kafedrasi assistenti
Shamsutdinova N.SH.
Annotation.Cell images, which have been widely used in biomedical research and drug discovery, contain a great deal of valuable information that encodes how cells respond to external stimuli and intentional perturbations. Meanwhile, to discover rarer phenotypes, cell imaging is frequently performed in a highcontent manner. Consequently, the manual interpretation of cell images becomes extremely inefficient. Thus, we also review more advanced machine learning technologies, aiming to make deep learningbased methods more useful and eventually promote the application of deep learning algorithms.
Key words: retraining module, DNN, machinelearning algorithms, CellCognition, inference module.
Cell image analysis plays an important role in biomedical research; it has become the main strategy for projecting how a life system interacts with environmental changes. For example, drug discovery is a crucial process for synthe sizing and screening potential candidate medications, which incurs a high cost to study the responses of intact cells or entire organisms to specific chemical substances. Phenotypic screening has been demonstrated to be a superior strategy for smallmolecule and firstinclass medicines based on phenotypic analysis of responses [1, 2]. Generally, to analyze the biological changes in cells, a phenotypic screening is per formed in a highcontent screening (HCS) method, where cells are stained with multichannel fluorescent probes and cultured in plates with multiple isolated walls subjected to different treatments [3–5]. Before entering clinical trials, molecules must be validated in vitro [6], which unfortunately has a high attrition rate. Moreover, previous research [7] indicates that using HCS to select small molecules only has an estimated hit rate of 0–0.01%, which is highly depen dent on the professional knowledge of different biologists and quality of the screening compound pool.
The goal of cell image analysis is to analyze the phenotypic effects of various treatments and to reveal the relation ships between them. The most widely studied tasks of cell image analysis include segmentation, tracking, and classifi cation [4]. These tasks have drawn extensive attention from both academia and industry. Recently, a bioimage challenge called the Cell Tracking Challenge for live cell seg mentation and tracking was held under the auspices of the
Notably, Recursion also released a series of opensource datasets called RxRx, aiming to extract phenotypic features by classifying the treatments imposed on cells based solely on raw image inputs. To make the quantitative and statistical analyses of cell images automated and high throughput, many software packages are available, such as ImageJ , CellProfiler , Icy , and CellCognition , which typically contain plentiful plugins that allow biologists to design a customized pipeline to perform differ ent tasks.
Deep learning, the most extensively used emerging machinelearning technique, has achieved remarkable suc cess in computer vision and natural language processing .In deep learning, a deep neural network (DNN) is trained as an endtoend model to directly infer the desired labels from input data . In contrast to traditional com puter vision techniques, a DNN can automatically produce more effective representations than handcrafted representa tions by learning from a largescale dataset. In cell images, deep learningbased methods also show promising results in cell segmentation and tracking . In addition, a vital breakthrough in computer vision called representa tion learning also provides confidence that pheno typic features can be learned endtoend by DNNs more efficiently.
Deep learning has shown a powerful ability to extract useful information from raw inputs; however, it is highly influenced by the quality of the dataset. As shown in Figure 1, a typical deeplearning method consists of two modules: an inference module and a retraining module. When the test environment changes, if the inference module does not achieve satisfactory generalization performance, the DNN must be retrained to adapt to the new data domain using extra annotations.
In this study, we provide a comprehensive survey of the current progress of three critical tasks. In addition, we willdiscuss the challenges of applying machinelearning algorithms to cell image analysis. In contrast to previous survey studies ,we provide a more technical perspective on deep learning in cell image analysis.
Second, region based methods require fewer computing resources than anchorbased methods do. This is because learning features with more “objectivity” often require a more complicated design and deeper neural networks. Third, regionbased methods require significantly more burdensome postprocessing than anchorbased methods. Deep learning based methods can achieve remarkable performance in cell segmentation after training with a largescale and carefully annotated dataset. However, cell images can vary with different treatments, such as different cell types, stains, or even carbon dioxide concentrations. For wholecell segmentation, Cellpose proposed a generalized dataset containing 10 different treatments (the value is approximated because the Cellpose dataset contains images col lected from Google) and up to 608 images, with 69 images held out for testing. In particular, this dataset also included 184 noncell images shared with repeating convex patterns for better generalization. For more specific applications, Greenwald et al. built a largescale tissue fluorescence dataset called TissueNet across six imaging platforms and nine organs, which also covered different disease states and spe cies.
Tracking. Tracking is another fundamental task in cell image analysis. Monitoring cell behaviors throughout the lineage can provide useful information for drug discovery, including quantification of signaling dynamics , efforts to understand cell motility , and attempts to unravel the laws of bacterial cell growth . Such an analysis must associate each cell entity over time. However, simultaneously tracking thousands of cells is challenging.

Score

Figure 2: Comparison of anchor and regionbased methods.

Third, owing to the phototoxicity and photobleaching during imaging, the frame rate of imaging is often limited.
Thus, the tracking context was not implicitly involved in the network training. Classification. Classification often serves as a down stream analysis task for phenotypic screening and cell profil ing. After individual cells are located, each cell is conducted into a highdimensional feature vector that contains various types of phenotypic information. Typical applications include

1. 
   (b) (c)
   

Figure 3: Schematic of general uncertainty estimation methods. (a) Bayes by Backprop: consider the weights of the neural network as Gaussian distributions (denoted by the purple lines). (b) MC dropout: adopt dropout in the training and inference phase, where the gray circle indicates the neuron dropped by dropout. (c) Deep ensembles: assemble multiple neural networks (denoted by the stacked rectangles). neural networks are incapable of capturing uncertainty, which causes some unexpected problems, resulting in diffi culties